config.yml

feature_extraction:
   name: pyannote.audio.features.RawAudio
   params:
      sample_rate: 16000

data_augmentation:
   name: pyannote.audio.augmentation.noise.AddNoise
   params:
     snr_min: 5
     snr_max: 15
     collection:
       - MUSAN.Collection.BackgroundNoise
       - MUSAN.Collection.Music

architecture:
   name: pyannote.audio.models.SincTDNN
   params:
      sincnet:
         stride: [5, 1, 1]
         waveform_normalize: True
         instance_normalize: True
      tdnn:
         embedding_dim: 512
      embedding:
         batch_normalize: False
         unit_normalize: False

task:
   name: AdditiveAngularMarginLoss
   params:
      margin: 0.05
      s: 10
      duration: 2.0
      per_fold: 256
      per_label: 1
      per_epoch: 5
      per_turn: 1
      label_min_duration: 30

scheduler:
   name: ConstantScheduler
   params:
      learning_rate: 0.01